Inflect-Mini-v3 / extras /events /command.py
owensong's picture
Release eight vocal-event tags across all six Mini voices
dc6dc7d verified
Raw History Blame Contribute Delete
6.11 kB
"""Eight vocal-event tags across six voices. CPU, native 24 kHz."""
import sys
sys.dont_write_bytecode = True
import argparse
import hashlib
import json
from pathlib import Path, PurePosixPath
import types
CORE_SHA = '3df1dd36fe67562448c2e9560ae8d8610cd5a6369fd21f966bc002fab1308f95'
COMPONENT_SHA = '55ecf645c7aa2b28d2112a0d25e000f48a723e19638c8f5ba160fdc5e76f4045'
BANK_MANIFEST_SHA = '161cbf0d348323eef7018b1fd8f7264dc968726abe01da99eb6edb8ad555f9f0'
VOICES = ('cassidy', 'hope', 'sophia', 'james', 'jasper', 'mark')
TAGS = ('sigh', 'gasp', 'cough', 'laugh', 'throat_clear', 'inhale', 'exhale', 'yawn')
def sha(path):
with Path(path).open('rb') as stream:
return hashlib.file_digest(stream, 'sha256').hexdigest()
def verify_tree(root, manifest_name, expected):
raw_root = Path(root)
if raw_root.is_symlink():
raise ValueError('Linked release root')
root = raw_root.resolve(strict=True)
manifest = root / manifest_name
if manifest.is_symlink() or sha(manifest) != expected:
raise ValueError('Release manifest pin mismatch')
data = json.loads(manifest.read_text(encoding='utf-8'))
for name, digest in data['files'].items():
rel = PurePosixPath(name)
if rel.is_absolute() or '..' in rel.parts or '\\' in name or ':' in name:
raise ValueError('Unsafe release path')
path = root.joinpath(*rel.parts)
if (not path.is_file() or not path.resolve().is_relative_to(root)
or any(root.joinpath(*rel.parts[:i]).is_symlink() for i in range(1, len(rel.parts) + 1))
or sha(path) != digest):
raise ValueError('Changed release asset: ' + name)
return root
def module(path, name):
result = types.ModuleType(name)
result.__file__ = str(path)
exec(compile(path.read_bytes(), str(path), 'exec'), result.__dict__)
return result
def load(package_root, *, events_root=None, diagnostic=False, profile='base', frontend='core', laugh_edge_room_ms=0):
package = Path(package_root).resolve(strict=True)
here = Path(events_root).resolve(strict=True) if events_root else package / 'extras/events'
core = verify_tree(package / 'core', 'assembly-manifest.json', CORE_SHA)
component = verify_tree(package / 'extras/events/component', 'manifest.json', COMPONENT_SHA)
bank = verify_tree(here / 'multivoice', 'manifest.json', BANK_MANIFEST_SHA)
bridge = module(component / 'runtime.py', '_integrated_native_bridge')
sdk = module(component / 'sdk.py', '_integrated_shared_sdk')
speech = bridge.GeneralCoreSpeech(core, CORE_SHA, profile=profile, frontend=frontend)
base = sdk.ApprovedEvents(speech, component, COMPONENT_SHA, rng_lock=speech.rng_lock)
source = module(bank / 'runtime.py', '_integrated_bank_runtime')
shim = module(bank / 'adapter.py', '_integrated_bank_adapter')
adapter = shim.make_adapter(source, base, bank, sha(bank / 'bank.json'), diagnostic=diagnostic,
laugh_edge_room_ms=laugh_edge_room_ms)
return adapter
def main():
here = Path(__file__).resolve().parent
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--core-root', type=Path, default=here.parents[1] / 'core')
parser.add_argument('--component', type=Path, default=here / 'component')
parser.add_argument('--verify', action='store_true')
parser.add_argument('--text')
parser.add_argument('--output', type=Path)
parser.add_argument('--voice', choices=VOICES, default='cassidy')
parser.add_argument('--profile', choices=('base', 'pitch'), default='base')
parser.add_argument('--frontend', choices=('core', 'general'), default='core')
parser.add_argument('--events', action='store_true')
parser.add_argument('--seed', type=int, default=42)
parser.add_argument('--laugh-edge-room-ms', type=int, choices=(0, 20), default=0)
parser.add_argument('--subtype', choices=('PCM_16', 'FLOAT'), default='PCM_16')
parser.add_argument('--allow-unreviewed-private', action='store_true')
args = parser.parse_args()
package = args.core_root.resolve(strict=True).parent
if args.component.resolve(strict=True) != package / 'extras/events/component':
parser.error('Use the shared component belonging to this package')
verify_tree(args.core_root, 'assembly-manifest.json', CORE_SHA)
verify_tree(args.component, 'manifest.json', COMPONENT_SHA)
bank = verify_tree(here / 'multivoice', 'manifest.json', BANK_MANIFEST_SHA)
if args.verify:
print(json.dumps(dict(verified=True, voices=VOICES, tags=TAGS, owner_release_authorized=True,
bank_manifest_sha256=BANK_MANIFEST_SHA,
historical_level_review=json.loads((bank / 'bank.json').read_text())['level_calibration']['human_approved'])))
return
if not args.text or args.output is None or not 0 <= args.seed < 2**32 - 100:
parser.error('Text, new output, and supported seed required')
output = args.output.resolve()
if args.output.is_symlink() or output.exists() or not output.parent.is_dir() or output.is_relative_to(package) or output.is_relative_to(here):
parser.error('New output outside the entire package and staged event directory required')
adapter = load(package, events_root=here, diagnostic=args.allow_unreviewed_private,
profile=args.profile, frontend=args.frontend, laugh_edge_room_ms=args.laugh_edge_room_ms)
result = adapter.synthesize(args.text, voice=args.voice, seed=args.seed, events=args.events)
import soundfile as sf
with output.open('xb') as stream:
sf.write(stream, result['audio'].detach().cpu().numpy(), 24000, format='WAV', subtype=args.subtype)
print(json.dumps(dict(audio_sha256=sha(output), sample_rate=24000, metadata=result['metadata'],
diagnostic=args.allow_unreviewed_private, owner_release_authorized=True)))
if __name__ == '__main__':
main()